An extended abstract of an indepth algorithmic and computational study for maximum flow problems

For a number of years the maximum flow network problem has attracted the attention of prominent researchers in network optimization. Since the groundbreaking work of Ford and Fulkerson 133, a variety of algorithms featuring good “worst-case” bounds bave been proposed for this problem. Surprisingly though, there has been almost no empirical evaluations of these algorithms. Cheung [l] recently conducted the first significant computational investigation of maximum flow methods, testing several of the major approaches Although an important step in the right direction, Cheung’s implementations employ methodology and data structures originating at least a dozen years ago. In the past decade, however, advances in network implementation technology have been dramatic. Sophisticated labeling techniques and more effective data structures have (a) decreased total solution time and/or (b) reduced computer memory requirements. As a result, widely held beliefs about which algorithms are best for particular problem classes have been steadily challenged and in some cases completely overturned. This study described in this abstract, likewise, discloses several misconceptions about maximum flow algorithms whose challenge was overdue. One of the major purposes of this study, therefore, has been to